A bus-based fully automatic 3D reconstruction system for non-cooperative ship targets

Abstract

The present disclosure proposes a bus-based fully automatic three-dimensional reconstruction system for non-cooperative ship targets, which belongs to the field of image processing technology. The system includes: a system bus layer and a functional module layer, wherein the functional module layer contains multiple functional modules for realizing data processing at different stages of the three-dimensional reconstruction process of non-cooperative ship targets; the system bus layer is connected to each module in the functional module layer via a data I / O interface, which is used to call the selected module from the functional module layer to obtain the corresponding data for processing and store the corresponding processing results. The present disclosure introduces the design concept of bus technology. Through the design of modular plug-ins, it can expand the universality of the system, improve the flexibility of the system, and effectively realize the fully automatic three-dimensional reconstruction of reconnaissance images of non-cooperative ship targets on the sea surface.

Description

Technical Field

[0001] The present disclosure belongs to the field of image processing technology, and in particular relates to a bus-based fully automatic three-dimensional reconstruction system for non-cooperative ship targets. Background Art

[0002] With the increasing autonomy of early warning, reconnaissance, detection, tracking, and precision guidance equipment, there is a need for rapid and accurate identification of non-cooperative surface vessels based on visible light imagery. However, since multi-angle, multi-position image datasets alone cannot intuitively capture the depth information of such targets, the industry currently relies on manual 3D rendering or contact measurement for modeling, which is not only inefficient but also difficult to accurately grasp the actual size of non-cooperative vessel targets. Based on visible light reconnaissance images of non-cooperative vessel targets, 3D reconstruction of non-cooperative vessel targets is performed. A 3D model database of various ship types is established and loaded into various autonomous identification equipment. This allows for accurate identification of vessel targets from any angle, significantly improving the analysis of non-cooperative vessel targets.

[0003] 3D reconstruction technology based on multi-view visible light images primarily involves the structure-from-motion (SFM) method. It utilizes theories and key technologies from fields such as geometric optics, mathematical algorithms, anatomy, neurophysiology, computer technology, statistics, operations research, graph theory, and signal analysis and processing to construct 3D models of objects or scenes. Due to the complexity of the reconstruction process, the breadth of domains involved, and the diverse research directions, a large number of algorithms have been developed for each specific reconstruction step. Furthermore, most current mainstream 3D reconstruction software focuses on software frameworks and functional implementation. Not every software can handle all image data that meets 3D reconstruction standards. In other words, the reconstruction targets and scenes they handle are all highly specific. Therefore, how to select the optimal algorithm for each step in real time for image data from specific scenes containing non-cooperative ship targets, and how to effectively combine these algorithms to build a flexible, efficient, and fully automated 3D reconstruction system is a practical challenge. Summary of the Invention

[0004] This paper aims to address the problem of efficient 3D reconstruction of non-cooperative ship targets by proposing a bus-based, fully automated 3D reconstruction system for non-cooperative ship targets. This system incorporates the design concept of bus technology and, through a modular plug-in design, expands the system's universality and flexibility, effectively achieving fully automated 3D reconstruction of reconnaissance images of non-cooperative ship targets on the sea surface.

[0005] An embodiment of the present disclosure proposes a bus-based fully automatic three-dimensional reconstruction system for non-cooperative ship targets, the system comprising: a system bus layer and a functional module layer, wherein the functional module layer comprises a plurality of functional modules, which are divided into a pre-processing subsystem, an image stitching subsystem, a three-dimensional reconstruction subsystem and a model display subsystem, and is used to fully automatically realize data processing at different stages in the three-dimensional reconstruction process of non-cooperative ship targets, and can select one-click automatic processing or wizard-style processing mode, process large quantities of image data with / without POS, and support interactive addition of control points and measurement of ship three-dimensional models; the system bus layer is used to call the selected module from the functional module layer to obtain corresponding data for processing and store the corresponding processing results.

[0006] In one embodiment of the present disclosure, the system bus layer is connected to each module in the functional module layer through a data I / O interface function, and the system bus layer calls the algorithm module in the functional module layer by plugging and unplugging the interface.

[0007] In one embodiment of the present disclosure, the system bus layer includes: a data bus unit, an address bus unit and an overlapping timing control bus unit; the address bus unit is used to read data from each functional module in the functional module layer, and use the functional module layer to store processing results; the overlapping timing control bus unit is used to obtain data from each module in the functional module layer from the address bus unit through a data engine and transmit it to the data bus unit; the data bus unit is used to store data obtained from the functional module layer.

[0008] In one embodiment of the present disclosure, the functional module layer includes: an image preprocessing module, a feature point extraction module, a feature matching module, an aerial triangulation module, a dense matching module, a grid construction module, a texture mapping module, an orthophoto generation module, a tiling module, an image stitching module and a report generation module; as needed, different modules can be flexibly combined and interacted through a bus, and three-dimensional reconstruction can be performed with overlapping working sequence, with the functions of automatic comprehensive reconstruction, automatic three-dimensional reconstruction, and step-by-step three-dimensional reconstruction.

[0009] In one embodiment of the present disclosure, the image preprocessing module is used to preprocess the acquired non-cooperative ship target image;

[0010] The feature point extraction module is used to detect the feature points of ships under conditions of multi-view rotation, different scales, different lighting and contrast, etc. on the pre-processed non-cooperative ship target image, with high reproducibility;

[0011] The feature matching module uses a hash matching method to match the detected ship feature points and automatically reconstructs the index of the matching points, connects the pre-processed ship images into a unified model and performs adjustment, and is insensitive to noise;

[0012] The aerial triangulation module is used to determine the position and attitude of the camera according to the camera parameters calibrated by the matching points, restore the position of the points in three-dimensional space, form a sparse point cloud of the ship target, and make the measurement accuracy pixel reprojection error no more than 1 pixel through the joint adjustment method;

[0013] The dense matching module is used to generate a three-dimensional dense point cloud of the ship target based on the sparse point cloud in combination with an adaptive step size algorithm, so as to achieve high reconstruction accuracy and integrity with few noise points and outliers;

[0014] The mesh construction module is used to build a triangular mesh model of the target based on the dense point cloud by using a surface fitting method and a triangular mesh method, and automatically repair holes by using a mesh optimization algorithm;

[0015] The texture mapping module is used to select the optimal reference image for each triangle based on the triangular mesh model and the camera parameters, cluster the reference images, optimize and generate the texture image of the scene, and automatically generate the texture image and automatically equalize the color differences between the images through rendering optimization to obtain the optimized three-dimensional ship model. The rendering also adapts to factors such as lighting differences, reconstruction accuracy, and limited calibration accuracy.

[0016] The orthophoto image generation module is used to realize the two-dimensional projection of the three-dimensional ship model at any angle by using a projection method, thereby forming a mosaic image of the ship at any angle. The mosaic image can be previewed in real time and can be compared with the original image at the same angle for error;

[0017] The tiling module performs layered and graded LOD processing on the three-dimensional model to generate corresponding tile data, thereby enabling real-time browsing and network publishing of the three-dimensional ship model data, and can be quickly imported into a third-party geographic information system.

[0018] The image stitching module is used to stitch the image data of the entire shooting scene, including the ship target, using an image stitching algorithm based on the results of the aerial triangulation module to form a panoramic image. Through the feature point optimization method, it can eliminate weakly connected images, images separated from the aerial photography area, and images with exposure failure.

[0019] The report generation module is based on the program measurement instruction record and is used to generate corresponding evaluation reports for the results of the remaining functional modules in the algorithm functional module layer, which can determine the advantages and disadvantages of different algorithms and provide algorithm selection strategies.

[0020] The characteristics and beneficial effects of the present disclosure are:

[0021] 1. This disclosure places the focus of the 3D reconstruction process on the flow of data. In the process of 3D reconstruction based on multi-view visible light images of non-cooperative ship targets, sparse point clouds, dense point clouds, surface models, texture models, panoramic images, orthophotos and other products can be generated.

[0022] 2. This disclosure proposes a bus-based, fully automated 3D reconstruction system for non-cooperative ship targets, equipped with data scheduling and management capabilities, enabling convenient access to various data types during the 3D reconstruction process. Image databases and related software tools are utilized to store, query, and retrieve image data, model data, and intermediate results. Various data access engines are established, driving the flow of data between different functional modules. Interfaces between different modules and the database ensure smooth data connectivity.

[0023] 3. The bus-based, non-cooperative, fully automated 3D reconstruction system for ship targets proposed in this disclosure utilizes a modular design, offering excellent stability, ease of maintenance, and scalability. For algorithms with the same functionality but different performance, each module can be plugged and unplugged within the overall software system through a unified functional module data I / O interface, thereby increasing the flexibility of software function combinations. A unified algorithm function call interface allows for the invocation of algorithms with varying performance, further enhancing the flexibility of algorithm performance combinations.

[0024] 4. The bus-based non-cooperative ship target fully automatic three-dimensional reconstruction system proposed in this disclosure has the function of pre-processing sea reconnaissance images, and can handle problems such as target image distortion, cloud and fog obstruction, and dim color caused by factors such as complex sea surface environment, weather changes, and camera shake. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1 It is a structural diagram of a bus-based non-cooperative ship target fully automatic three-dimensional reconstruction system in an embodiment of the present disclosure.

[0026] Figure 2 It is a schematic diagram of the system bus layer structure in an embodiment of the present disclosure.

[0027] Figure 3 It is a design diagram of a single functional module architecture in an embodiment of the present disclosure.

[0028] Figure 4 It is an architectural principle diagram of the functional module layer in the embodiment of the present disclosure.

[0029] Figure 5This is a working sequence diagram of a bus-based non-cooperative ship target fully automatic three-dimensional reconstruction system in an embodiment of the present disclosure.

[0030] Figure 6 This is a workflow diagram of a bus-based non-cooperative ship target fully automatic three-dimensional reconstruction system in an embodiment of the present disclosure. DETAILED DESCRIPTION

[0031] The following will be combined with the accompanying drawings in the embodiments of the present disclosure to clearly and completely describe the technical solutions in the embodiments of the present disclosure. Based on the embodiments of the present disclosure, all other embodiments obtained by technicians in the relevant fields without making creative work are within the scope of protection of the present disclosure.

[0032] Those skilled in the relevant art should understand that the following embodiments are merely used to explain the technical principles of the present disclosure and are not intended to limit the scope of protection of the present disclosure.

[0033] The embodiment of the present disclosure proposes a bus-based non-cooperative ship target fully automatic 3D reconstruction system, the structure of which is as follows: Figure 1 As shown, the system includes a system bus layer and a functional module layer. The functional module layer contains multiple functional modules for implementing data processing at different stages of the 3D reconstruction of non-cooperative ship targets. The system bus layer is used to retrieve selected module combinations from the functional module layer to obtain corresponding data for processing and store the corresponding processing results. The system bus layer is connected to each module in the functional module layer via a standard data I / O interface. The system bus layer retrieves the module combinations in the functional module layer by plugging and unplugging these interfaces.

[0034] In some embodiments of the present disclosure, the functional module layer includes: an image preprocessing module, a feature point extraction module, a feature matching module, an aerial triangulation module, a dense matching module, a grid construction module, a texture mapping module, an orthophoto generation module, a tiling module, an image stitching module and a report generation module.

[0035] The system bus layer structure is as follows Figure 2As shown, it includes: a data bus unit, an address bus unit and an overlapping timing control bus unit. Among them, the data bus unit is used to store image data and result data. The unit contains: original image database, preprocessing result library, feature point detection result library, stereo matching result library, image stitching library, dense point cloud and sparse point cloud library and other databases; the overlapping timing control bus unit is used to establish data, storage identification, location, and perform data storage and reading. Various data flows can be freely called according to different interfaces. The unit contains: original data image reading engine, preprocessing result storage engine, feature point detection result reading engine, stereo matching reading result engine, dense point cloud and sparse point cloud storage engine, as well as aerial triangulation result reading storage engine and other multiple Engine; the address bus unit is used to read the required data of each database in the data bus unit from each module of the functional module layer through the I / O interface, and use the functional modules of the functional module layer to store the corresponding processing results; each functional module is used to obtain data from the database of the data bus unit through the address bus unit using the I / O interface, thereby realizing the docking of the data in the database of the data bus unit with the functional module; the architecture of the functional modules in the embodiment of the present disclosure is based on the bus technology concept, which can realize the plug-in and combination of different functions, and can select one-click automatic processing or wizard-style processing mode, process large quantities of image data with / without POS (position and orientation system), and support interactive addition of control points and measurement of three-dimensional models; the overall architecture design of each functional module is as follows Figure 3 As shown in the figure, the algorithm calling interface of the functional module needs to be defined as a standardized interface for the algorithm of the relevant function, so as to call different algorithm dynamic libraries and realize the pluggability of different algorithms in the software sense, thereby improving the adaptability of the system to the target scenario; the data I / O interface of the functional module is connected with the corresponding interface in the address bus to call different functional modules and realize the pluggability of different functional modules in the software sense;

[0036] The standardized interfaces in the disclosed embodiments are mainly divided into two categories: one is the data I / O storage format standard defined by the database access operation in the data bus, and the other is the function interface standard defined by the algorithm call operation in the function module;

[0037] In the embodiment of the present disclosure, the image preprocessing module mainly performs haze removal, geometric correction, light and color uniformity, and color enhancement on the input maritime reconnaissance image. By integrating relevant algorithms, the processed image can be made clearer, with more vivid colors, corrected distortion, and more uniform brightness.

[0038] In the embodiment of the present disclosure, the feature point extraction module mainly uses mainstream algorithms such as Harris, SIFT, and SURF to detect feature points of the pre-processed non-cooperative ship target image under conditions such as multi-view rotation, different scales, different lighting and contrast, so that the detected corner points or spots have high reproducibility, positioning accuracy and detection efficiency;

[0039] In the disclosed embodiment, the feature matching module mainly performs feature point matching between images based on the hash matching method, and automatically reconstructs the matching point index, connects the discrete pre-processed images into a unified model and performs adjustment; and achieves insensitivity to noise and small matching errors through the matching optimization algorithm;

[0040] In the embodiment of the present disclosure, the aerial triangulation module mainly uses the SFM method to determine the position and attitude of the camera based on the camera parameters calibrated by the matching points, and then restores the position of the points in three-dimensional space to form a sparse three-dimensional structure of the target; by performing a joint adjustment method on a large number of pre-processed images, the measurement accuracy pixel reprojection error (mean square error) does not exceed 1 pixel;

[0041] In the disclosed embodiment, the dense matching module generates a three-dimensional dense point cloud of a non-cooperative ship target based on a sparse point cloud, and based on the CMVS and PMVS2 calculation modules, thereby reconstructing the detailed information of the ship. By using an adaptive step size algorithm, the reconstructed information has high accuracy and integrity, with few noise points and outliers, and the time and space complexity of the reconstruction algorithm are simplified.

[0042] In the disclosed embodiment, the mesh construction module mainly builds the target's triangular mesh structure based on dense point clouds through surface fitting and triangular meshing methods to reconstruct the target's surface model. Through the mesh optimization algorithm, the mesh construction has high accuracy, is insensitive to noise, and can automatically repair holes.

[0043] In the disclosed embodiment, the texture mapping module is designed to select the optimal reference image for each triangle based on the reconstructed triangular mesh model and calibrated camera parameters, cluster the reference images of the triangular faces, and optimize them to generate a texture image of the scene, thereby obtaining a 3D model with texture visualization and realism. Through rendering optimization, the texture image can be fully automatically generated, the color differences between images can be fully automatically balanced, and no obvious texture seams can be achieved. The rendering can also adapt to factors such as lighting differences, reconstruction accuracy, and limited calibration accuracy.

[0044] In the embodiment of the present disclosure, the orthoimage generation module uses a projection method to realize two-dimensional projection of the target model at any angle, forming a mosaic image at any angle; the mosaic image can be previewed in real time, a pyramid can be quickly generated, and the error can be compared with the original image at the same angle as a basis for adjusting the model accuracy;

[0045] In the disclosed embodiment, the tiling module performs level of detail (LOD) processing on the 3D ship model to generate tile data of different precisions, enabling real-time browsing and network publishing of the 3D model data. The 3D model can also be browsed and displayed in real time on a common office PC and can be quickly imported into a third-party geographic information system (GIS).

[0046] In the disclosed embodiment, the image stitching module uses an image stitching algorithm based on the results of aerial triangulation to stitch the image data of the entire shooting scene, including the ship target, to form a panoramic image, thereby achieving an overview of the target scene. The feature point optimization method can eliminate weakly connected images, images separated from aerial photography areas, and images with exposure failure.

[0047] In the embodiment of the present disclosure, the report generation module generates an objective accuracy evaluation report for the results of the remaining functional modules in the functional module layer based on the program measurement instruction records. Therefore, under the condition of the same input data, when the same module calls different algorithms, the advantages and disadvantages of the relevant algorithms can be determined based on the generated evaluation report, and then the algorithm selection strategy for different goals can be given.

[0048] The working principle of the system of the embodiment of the present disclosure is as follows:

[0049] The functional module layer of the bus-based non-cooperative ship target fully automatic 3D reconstruction system of the embodiment of the present disclosure can be divided into four subsystems: Figure 4As shown, the system includes a preprocessing subsystem, an image stitching subsystem, a 3D reconstruction subsystem, and a model display subsystem. The preprocessing subsystem is responsible for sea surface separation, filtering out sea surface information and extracting the fine outlines of ships in images of ships on the sea surface. Image stitching is responsible for generating panoramic images. The 3D reconstruction subsystem includes three reconstruction modes: automatic 3D reconstruction, automatic integrated reconstruction, and step-by-step 3D reconstruction. The automatic 3D reconstruction mode, based on a multi-view environment, enables fully automatic reconstruction of small near-field objects and large distant scenes. The automatic integrated reconstruction mode, in addition to 3D reconstruction, also features tiling of 3D model data and can generate DEMs and orthophoto mosaics. The step-by-step 3D reconstruction allows for the step-by-step reconstruction of small near-field objects and can read intermediate-level products output by other software, such as sparse point clouds, dense point clouds, and mesh models, for subsequent processing. The model display subsystem allows for viewing the point cloud models, mesh models, surface models, and texture models in the reconstruction results.

[0050] In a specific embodiment of the present disclosure, the functional module layer is divided into the following according to different functions: Figure 1 The different functional modules shown, wherein the reference of the basic storage format standard of data I / O of each functional module is shown in Table 1.

[0051] Table 1 Reference table of basic storage format standards for functional module data I / O in the embodiment of the present disclosure

[0052]

[0053]

[0054] Among them, the specific data element arrangement and storage standards in each data storage file comply with relevant international common practices.

[0055] Throughout the 3D reconstruction process, each functional module has strong inheritance. For example, from feature point detection to texture mapping, the next module cannot operate without the results of the previous module. Therefore, when defining each functional module to call the same function but different performance algorithms, a unified function interface must be used. Taking C++ language as an example, the pseudo code structure is:

[0056] bool function interface name (input data header, input configuration parameter header, output data header, output configuration, parameter header); if the return value is 1, it indicates success; if the return value is 0, it indicates failure.

[0057] During the working process of the system, for the connection between each functional module of the fully automatic 3D reconstruction system for non-cooperative ship targets and each database in the data bus unit, the overlapping timing control bus unit controls each access engine as the hub of the data flow between the two. The specific working timing design is as follows: Figure 5 As shown, for example, an image is read from the overlapping timing control bus unit and sent to the image preprocessing module. The preprocessing results are then stored in the corresponding database in the data bus unit. This shows that the overlapping timing control bus unit controls both the storage engines and the functional modules. It's important to note that from the start of the feature detection module to the completion of the orthoimage generation module, the operating timing of each module overlaps to some extent. This development is designed to accelerate system processing speed and improve processor and memory utilization.

[0058] The operation process of the bus-based non-cooperative ship target fully automatic 3D reconstruction system according to the embodiment of the present disclosure is as follows: Figure 6As shown. After loading the image to be reconstructed, the image group window displays the image list; the operation history window records the commands for calling executable files or dynamic libraries, operation parameter settings and other information; the run record window records the key steps of the reconstruction process by time; the model viewing window can check whether the reconstruction results meet the requirements, and can rotate, zoom and other operations through the coordinate team model; when creating a new project, you can select the automatic 3D reconstruction, comprehensive 3D reconstruction and step-by-step 3D reconstruction mode in the project by selecting the project mode drop-down menu; when opening a project, you can select the existing project folder path and select the reconstruction mode used in the existing project through the drop-down menu; in the pre-processing menu item, set the significant area extraction parameters. Taking the separation of sea surface function as an example, select the image path to be processed and the save path of the processed image respectively. The rough contour Gaussian variance, pyramid Gaussian variance and fine contour Gaussian variance in the middle can be set according to different images, or not set to be automatically processed by the corresponding algorithm; the automatic 3D reconstruction item can set the automatic 3D reconstruction parameters. The larger the tree depth value, the finer the model, the smaller the screening threshold, and the more parts of the model are filtered out; the automatic comprehensive reconstruction item sets the automatic comprehensive reconstruction parameters and creates the necessary data preparation The reconstruction scale parameter sets the image downsampling factor. The texture file size sets the size of the texture patch file and can be adjusted based on the image data size. The texture mapping downsampling parameter sets the resolution reduction factor for the patch. You can choose whether to perform LOD blocking based on different scenarios. The sparse reconstruction function of the step-by-step 3D reconstruction function allows you to select different feature point detection methods and set a maximum resolution limit in the feature extraction method. If you want to reduce the image resolution, you can check the Set Reduced Resolution Sparse option and select the corresponding factor. The dense matching function of the step-by-step 3D reconstruction function requires you to select the path for the sparse reconstruction results and set the output path for the dense matching results. The surface reconstruction function of the step-by-step 3D reconstruction function requires you to select a dense point cloud ply format file and set the output path for the reconstructed mesh results. A larger tree depth parameter results in a finer model, and a larger screening threshold value results in more redundant parts being filtered out. The texture mapping function of the step-by-step 3D reconstruction function requires you to select the output path for the dense matching results and set the texture mapping result output path. You can also select the ply format file for the surface reconstruction output. The visualization module is set to include four options under the 3D view menu: sparse point cloud, dense point cloud, mesh model, and texture model, allowing you to view the corresponding output results in real time.

[0059] Heretofore, typical embodiments of the present disclosure have been described with reference to the accompanying drawings.

[0060] Although the various steps are described in chronological order in the above embodiments, those skilled in the art will appreciate that, in order to achieve better results in this embodiment, the execution order of the local steps may be adjusted appropriately, and the steps may be executed simultaneously (in parallel). These simple changes are within the scope of protection of this disclosure.

[0061] The method steps disclosed herein can be implemented using electronic hardware, computer software, or a combination of both, with the specific implementation depending on the specific application and design constraints of the technical solution.

[0062] For technicians in the relevant fields, the protection scope of the present disclosure is obviously not limited to the above-mentioned specific implementation methods. Without departing from the principles of the present disclosure, any modification, replacement and application of the bus-based non-cooperative ship target fully automatic three-dimensional reconstruction system falls within the protection scope of the present disclosure.

Claims

1. A bus-based non-cooperative ship target fully automatic 3D reconstruction system, characterized by: The system comprises: a system bus layer and a functional module layer, wherein the functional module layer comprises: an image preprocessing module, a feature point extraction module, a feature matching module, an aerial triangulation module, a dense matching module, a grid construction module, a texture mapping module, an orthophoto generation module, a tiling module, an image stitching module and a report generation module; the system bus layer is used to call selected modules from the functional module layer to obtain corresponding data for processing and store corresponding processing results; The image preprocessing module is used to preprocess the acquired non-cooperative ship target image; The feature point extraction module is used to detect ship feature points on the pre-processed non-cooperative ship target image under multi-view rotation, different scales, different lighting and contrast conditions, with high reproducibility; The feature matching module uses a hash matching method to match the detected ship feature points and automatically reconstructs the index of the matching points, connects the pre-processed ship images into a unified model and performs adjustment, and is insensitive to noise; The aerial triangulation module is used to determine the position and attitude of the camera according to the camera parameters calibrated by the matching points, restore the position of the points in three-dimensional space, form a sparse point cloud of the ship target, and make the measurement accuracy pixel reprojection error no more than 1 pixel through the joint adjustment method; The dense matching module is used to generate a three-dimensional dense point cloud of the ship target based on the sparse point cloud in combination with an adaptive step size algorithm, so as to achieve high reconstruction accuracy and integrity with few noise points and outliers; The mesh construction module is used to build a triangular mesh model of the ship target based on the dense point cloud by using a surface fitting method and a triangular mesh method, and automatically repair holes by using a mesh optimization algorithm; The texture mapping module is used to select the optimal reference image for each triangle based on the triangular mesh model and the camera parameters, cluster the reference images, optimize and generate the texture image of the scene, and automatically generate the texture image and automatically equalize the color differences between the images through rendering optimization to obtain the optimized three-dimensional ship model. The rendering also adapts to factors such as lighting differences, reconstruction accuracy, and limited calibration accuracy. The orthophoto image generation module is used to realize the two-dimensional projection of the three-dimensional ship model at any angle by using a projection method, thereby forming a mosaic image of the ship at any angle. The mosaic image can be previewed in real time and can be compared with the original image at the same angle for error; The tiling module performs layered and graded LOD processing on the three-dimensional model to generate corresponding tile data, thereby enabling real-time browsing and network publishing of the three-dimensional ship model data, and can be quickly imported into a third-party geographic information system. The image stitching module is used to stitch the image data of the entire shooting scene, including the ship target, using an image stitching algorithm based on the results of the aerial triangulation module to form a panoramic image. Through the feature point optimization method, weakly connected images, images separated from the aerial photography area, and images with exposure failure can be eliminated; The report generation module is based on the program measurement instruction record and is used to generate corresponding evaluation reports for the results of the remaining functional modules in the algorithm functional module layer, which can determine the advantages and disadvantages of different algorithms and provide algorithm selection strategies.

2. The system according to claim 1, wherein: The system bus layer is connected to each module in the functional module layer through a data I / O interface function, and the system bus layer can call the algorithm module in the functional module layer by plugging and unplugging the interface.

3. The system according to claim 2, characterized in that The system bus layer includes: a data bus unit, an address bus unit and an overlapping timing control bus unit; the address bus unit is used to read data from each functional module in the functional module layer and use the functional module layer to store processing results; the overlapping timing control bus unit is used to obtain data from each module in the functional module layer from the address bus unit through a data engine and transmit it to the data bus unit; the data bus unit is used to store data obtained from the functional module layer.

4. The system according to claim 2, wherein: Different modules in the functional module layer can be flexibly combined and interacted through a bus, and three-dimensional reconstruction can be performed with overlapping working sequences, with the functions of automatic comprehensive reconstruction, automatic three-dimensional reconstruction, and step-by-step three-dimensional reconstruction.

Images